Can the new IBM compilers be used to speed folding?

[Dros]

How hard would it be to recompile the clients with the new IBM compilers?

Macrumors thread

A 30%-40% speed gain would help the Mac teams quite a bit!

 

[MrMacMan]

Doubtful.

Anyway no one (I'm probably wrong) in the public has the source code for Folding... Its not Open Source, if it was client development would be a loottt faster.

 

[Dros]

Originally posted by MrMacman
Doubtful.

Anyway no one (I'm probably wrong) in the public has the source code for Folding... Its not Open Source, if it was client development would be a loottt faster.

So, doubtful because there would be no speed gains if anyone did compile it or because it will be difficult to convince someone to recompile it?

I guess if the new core is sitting in Altivec for most of its execution then it would depend how well the compilers optimize that part of the chip, so maybe not worth it. But if it would allow a significant speed-up and it is just a matter of setting a few flags...

 

[MrMacMan]

I really don't know.

I mean for anyone with a G5 ... how much boost would you get?

10% ?

Its impossible to know.

 

[bousozoku]

Stanford was looking at the IBM compilers as an option. The speed of the Tinker core could probably be enhanced quite a lot and since there's a new FORTRAN compiler, hopefully they will run more than a few tests. (They already had the betas last month.)

The Gromacs code is quite good now and it would not be the greatest need at this point. As well, the clients themselves will gain little from speedups, although the graphical clients would gain more than the console client would.